The contribution of High Performance Computing and Modelling for Industrial
Development
Dr Happy Sithole and Dr Onno Ubbink
Strategic context
• High-performance computing (HPC) combined with machine Learning and artificial intelligence present opportunities to non-traditional industries.
• The CSIR’s renewed focus on industrialisation opens up new opportunities for HPC adoption through virtual prototyping.
• Most of the developed countries have taken HPC as part of their competitiveness strategy.
Value proposition for industry
• CHPC experience of providing support to industry.
• We treat each client differently and recognise unique requirements.
• We work closely with each client to provide satisfactory solutions.
• We help to reduce the barrier for innovation through the “Virtual” Prototypes.
• Help move ideas from “Valley of Death” to realisation.
• We improve your risk management for innovation.
• Engineer for resilience.
Resource requirement
Value proposition for industry
What do we offer?
• Streamlined contracting model to fast track the process
• Experience Users can have access to CPU cycles and applications support
• Benchmarking of hardware and software and provide advice of optimal systems
• Commissioning of computing equipment, optimised for client’s needs
• A complete solution for the client where there is no capacity
• A gradual on-ramp where there is need to develop capacity in long term
Leadership HPC systems
System configuration Phase 1 Phase 2
Dell PowerEdge C6320 Servers:
Standard Compute nodes 128GB (64GB) / node 1 008 1 368
2 x Intel Xeon E5-2690 v3 (Haswell) processors (12 Cores Each 24 cores / node)
24 192 32 832
Dell PowerEdge R930 servers:
Large Memory Compute Nodes 1024GB / node (FAT nodes) 5 5 4 x Intel Xeon E7-4850 v3 processors
(14 Cores Each 56 cores / node)
280 280
Infiniband FDR 2:1 Blocking (56 Gbps)
Parallel Storage (Useable) PB 4 4
Total Number of Racks (including Compute, Login, Management and Storage Nodes)
19 24
Centos 7.1 with Bright Cluster Manager and Altair PBS Pro
Total Linpack Performance (Tflop/s) 783 1029
HPC Systems
• 127 on TOP500
• Awarded the fastest
supercomputer in Africa
Resource utilisation
Nov/Dec 2016
Jan/Feb 2017
Resource utilisation
Resource utilisation by domains
What happens during a design process?
Design it
Build it
Test it
Technology development continuum
Theoretical
&
basic research
Commercialisation
&
production (.co.za or .org) Virtual
prototyping
&
development
Optimisation
&
robustification
HPC
Bridges the Gap BETWEEN
Basic research & commercialisation
Product Life Cycle
Phase 1 Feasibility
Phase 3 Prototyping
Phase 4 Production/
Deployment Phase 0
Concept/
Vision
Phase 2 Design/
Development
Over the horizon development …
Visualisation Dashboard Information-
based decisions
Automated provenance Human,
physical, virtual sensors
Archives
Modelling Multimodal
Integrated, information-based
decision making
CSIR offerings
CSIR capabilities Defense Climate modelling
Materials Fluids
HPC systems
CHPC
applications and support
Market
SW Dev.
Modelling capabilities
Road dynamics
Vehicle-tyre-pavement interaction
Road dynamics
Fluid-structure interaction (FSI)
• Interaction of some movable or deformable structure with an internal or surrounding fluid flow
• a crucial consideration in the design of many engineered systems
Non-rigid surface (elastic) registration
• Process for determining the
correspondence of features between images:
• Collected at different times
• Captured using different imaging modalities
• Similar objects to be compared
Generic and target shape Selective feature registration
Variable-resolution Earth System Model (VRESM)
• As demonstrated by CCAM (world’s first cubic atmospheric model), the cubic grid approach provides
considerable advantages in numerical efficiency; this numerical efficiency allows finer resolution, longer
simulations and more ensemble members
• VRESM inherits CCAM’s reversible staggering grid that provides the model with its excellent wave-dispersive
properties
• VRESM also inherits CCAM’s extensive
Industry partners
Proof of concept and realisation of value
PetroSA Eclipse Benchmark Case
Dimensions 149X118X128 – 2.25 Million Cells
